From a645c4fd7ad1b37534f673eecdb5d8756c3b97bd Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Fri, 5 Jul 2019 16:12:11 +0200
Subject: [PATCH 01/21] FIX normalize with max of samples with non-null weights
 in AdaBoostRegressor

---
 doc/whats_new/v0.22.rst                       |  4 ++
 .../ensemble/tests/test_weight_boosting.py    | 39 ++++++++++++-------
 sklearn/ensemble/weight_boosting.py           | 20 +++++-----
 3 files changed, 40 insertions(+), 23 deletions(-)

diff --git a/doc/whats_new/v0.22.rst b/doc/whats_new/v0.22.rst
index f0670e1293369..9eff6ecf86d6c 100644
--- a/doc/whats_new/v0.22.rst
+++ b/doc/whats_new/v0.22.rst
@@ -60,6 +60,10 @@ Changelog
   parameter called `warm_start` that enables warm starting. :pr:`14012` by
   :user:`Johann Faouzi <johannfaouzi>`.
 
+- |Fix| :class:`ensemble.AdaBoostRegressor` where the loss should be normalized
+  by the max of the samples with non-null weights.
+  :pr:`xxx` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 :mod:`sklearn.linear_model`
 ...........................
 
diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 1cb1e9d1431cf..c7f47be2a9903 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -1,12 +1,16 @@
 """Testing for the boost module (sklearn.ensemble.boost)."""
 
 import numpy as np
+import pytest
 
 from sklearn.utils.testing import assert_array_equal, assert_array_less
-from sklearn.utils.testing import assert_array_almost_equal
+from sklearn.utils.testing import assert_array_almost_equal, assert_allclose
 from sklearn.utils.testing import assert_raises, assert_raises_regexp
+from sklearn.utils.testing import set_random_state
 
 from sklearn.base import BaseEstimator
+from sklearn.base import clone
+from sklearn.dummy import DummyClassifier, DummyRegressor
 from sklearn.model_selection import train_test_split
 from sklearn.model_selection import GridSearchCV
 from sklearn.ensemble import AdaBoostClassifier
@@ -304,16 +308,6 @@ def test_base_estimator():
                          clf.fit, X_fail, y_fail)
 
 
-def test_sample_weight_missing():
-    from sklearn.cluster import KMeans
-
-    clf = AdaBoostClassifier(KMeans(), algorithm="SAMME")
-    assert_raises(ValueError, clf.fit, X, y_regr)
-
-    clf = AdaBoostRegressor(KMeans())
-    assert_raises(ValueError, clf.fit, X, y_regr)
-
-
 def test_sparse_classification():
     # Check classification with sparse input.
 
@@ -486,9 +480,6 @@ def test_multidimensional_X():
     Check that the AdaBoost estimators can work with n-dimensional
     data matrix
     """
-
-    from sklearn.dummy import DummyClassifier, DummyRegressor
-
     rng = np.random.RandomState(0)
 
     X = rng.randn(50, 3, 3)
@@ -503,3 +494,23 @@ def test_multidimensional_X():
     boost = AdaBoostRegressor(DummyRegressor())
     boost.fit(X, yr)
     boost.predict(X)
+
+
+class ClassifierWithoutWeight(DummyClassifier):
+    """Classifier not supporting `sample_weight`."""
+    def fit(self, X, y):
+        super().fit(X, y)
+        return self
+
+
+@pytest.mark.parametrize("algorithm", ['SAMME', 'SAMME.R'])
+def test_adaboostclassifier_without_sample_weight(algorithm):
+    X, y = iris.data, iris.target
+    base_estimator = ClassifierWithoutWeight()
+    clf = AdaBoostClassifier(
+        base_estimator=base_estimator, algorithm=algorithm
+    )
+    err_msg = ("{} doesn't support sample_weight"
+               .format(base_estimator.__class__.__name__))
+    with pytest.raises(ValueError, match=err_msg):
+        clf.fit(X, y)
diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index 3cb4baa0d9a0c..09fc632421546 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -1039,13 +1039,10 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
         estimator = self._make_estimator(random_state=random_state)
 
         # Weighted sampling of the training set with replacement
-        # For NumPy >= 1.7.0 use np.random.choice
-        cdf = stable_cumsum(sample_weight)
-        cdf /= cdf[-1]
-        uniform_samples = random_state.random_sample(_num_samples(X))
-        bootstrap_idx = cdf.searchsorted(uniform_samples, side='right')
-        # searchsorted returns a scalar
-        bootstrap_idx = np.array(bootstrap_idx, copy=False)
+        bootstrap_idx = random_state.choice(
+            np.arange(_num_samples(X)), size=_num_samples(X), replace=True,
+            p=sample_weight
+        )
 
         # Fit on the bootstrapped sample and obtain a prediction
         # for all samples in the training set
@@ -1055,9 +1052,14 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
         y_predict = estimator.predict(X)
 
         error_vect = np.abs(y_predict - y)
-        error_max = error_vect.max()
 
-        if error_max != 0.:
+        sample_mask = sample_weight > 0
+        if not np.count_nonzero(sample_mask):
+            error_max = 0
+        else:
+            error_max = error_vect[sample_mask].max()
+
+        if error_max != 0:
             error_vect /= error_max
 
         if self.loss == 'square':

From cb52f6585271a8ad4a3ef66897f7f1bec555526e Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Mon, 8 Jul 2019 15:50:41 +0200
Subject: [PATCH 02/21] iter

---
 .../ensemble/tests/test_weight_boosting.py    | 39 ++++++++++++++++---
 sklearn/ensemble/weight_boosting.py           |  9 +++--
 2 files changed, 39 insertions(+), 9 deletions(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index c7f47be2a9903..7eecf03c4a884 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -3,6 +3,12 @@
 import numpy as np
 import pytest
 
+from scipy.sparse import csc_matrix
+from scipy.sparse import csr_matrix
+from scipy.sparse import coo_matrix
+from scipy.sparse import dok_matrix
+from scipy.sparse import lil_matrix
+
 from sklearn.utils.testing import assert_array_equal, assert_array_less
 from sklearn.utils.testing import assert_array_almost_equal, assert_allclose
 from sklearn.utils.testing import assert_raises, assert_raises_regexp
@@ -16,11 +22,7 @@
 from sklearn.ensemble import AdaBoostClassifier
 from sklearn.ensemble import AdaBoostRegressor
 from sklearn.ensemble import weight_boosting
-from scipy.sparse import csc_matrix
-from scipy.sparse import csr_matrix
-from scipy.sparse import coo_matrix
-from scipy.sparse import dok_matrix
-from scipy.sparse import lil_matrix
+from sklearn.linear_model import LinearRegression
 from sklearn.svm import SVC, SVR
 from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
 from sklearn.utils import shuffle
@@ -514,3 +516,30 @@ def test_adaboostclassifier_without_sample_weight(algorithm):
                .format(base_estimator.__class__.__name__))
     with pytest.raises(ValueError, match=err_msg):
         clf.fit(X, y)
+
+
+def test_adaboost_regressor_sample_weight():
+    # check that giving weight will have an influence on the error computed
+    # for a weak learner
+    X, y = datasets.make_regression(n_features=50, random_state=0)
+
+    # add an arbitrary outlier to make sure
+    X = np.vstack([X, X.sum(axis=0)])
+    y = np.hstack([y, 10])
+
+    regr_no_outlier = AdaBoostRegressor(
+        base_estimator=LinearRegression(), n_estimators=4, random_state=0
+    )
+    regr_with_weight = clone(regr_no_outlier)
+
+    # fit 2 models:
+    # - a model without the outlier
+    # - a model containing the outlier but with a null sample-weight
+    # Therefore, the error of the first weak learner will be identical.
+    regr_no_outlier.fit(X[:-1], y[:-1])
+    sample_weight = np.array([1.] * (y.size - 1) + [0.])
+    regr_with_weight.fit(X, y, sample_weight=sample_weight)
+
+    # check that the error is similar with 2 decimals
+    assert (regr_no_outlier.estimator_errors_[0] ==
+            pytest.approx(regr_with_weight.estimator_errors_[0], abs=1e-2))
diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index 09fc632421546..c1864ff6e48d6 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -1054,7 +1054,7 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
         error_vect = np.abs(y_predict - y)
 
         sample_mask = sample_weight > 0
-        if not np.count_nonzero(sample_mask):
+        if np.all(~sample_mask):
             error_max = 0
         else:
             error_max = error_vect[sample_mask].max()
@@ -1068,7 +1068,8 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
             error_vect = 1. - np.exp(- error_vect)
 
         # Calculate the average loss
-        estimator_error = (sample_weight * error_vect).sum()
+        estimator_error = (sample_weight[sample_mask] *
+                           error_vect[sample_mask]).sum()
 
         if estimator_error <= 0:
             # Stop if fit is perfect
@@ -1086,9 +1087,9 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
         estimator_weight = self.learning_rate * np.log(1. / beta)
 
         if not iboost == self.n_estimators - 1:
-            sample_weight *= np.power(
+            sample_weight[sample_mask] *= np.power(
                 beta,
-                (1. - error_vect) * self.learning_rate)
+                (1. - error_vect[sample_mask]) * self.learning_rate)
 
         return sample_weight, estimator_weight, estimator_error
 

From 805a6b51c716a067515931b8d2fefa3b65b5afc1 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Mon, 8 Jul 2019 15:54:13 +0200
Subject: [PATCH 03/21] update PR number

---
 doc/whats_new/v0.22.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/whats_new/v0.22.rst b/doc/whats_new/v0.22.rst
index 0345b65a260ba..3a14d275ea0b6 100644
--- a/doc/whats_new/v0.22.rst
+++ b/doc/whats_new/v0.22.rst
@@ -67,7 +67,7 @@ Changelog
 
 - |Fix| :class:`ensemble.AdaBoostRegressor` where the loss should be normalized
   by the max of the samples with non-null weights.
-  :pr:`xxx` by :user:`Guillaume Lemaitre <glemaitre>`.
+  :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
 
 :mod:`sklearn.linear_model`
 ...........................

From b0f2966d405bf855a1094675a22c1d3ca9d0d1ef Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Mon, 8 Jul 2019 15:54:43 +0200
Subject: [PATCH 04/21] PEP8

---
 sklearn/ensemble/tests/test_weight_boosting.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 7eecf03c4a884..833ac0091837c 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -10,9 +10,8 @@
 from scipy.sparse import lil_matrix
 
 from sklearn.utils.testing import assert_array_equal, assert_array_less
-from sklearn.utils.testing import assert_array_almost_equal, assert_allclose
+from sklearn.utils.testing import assert_array_almost_equal
 from sklearn.utils.testing import assert_raises, assert_raises_regexp
-from sklearn.utils.testing import set_random_state
 
 from sklearn.base import BaseEstimator
 from sklearn.base import clone

From 0f317d988f06881697cd1ebd7682727d30115bcc Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Mon, 8 Jul 2019 16:15:39 +0200
Subject: [PATCH 05/21] iter

---
 sklearn/ensemble/tests/test_weight_boosting.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 833ac0091837c..46992c9d19e07 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -539,6 +539,6 @@ def test_adaboost_regressor_sample_weight():
     sample_weight = np.array([1.] * (y.size - 1) + [0.])
     regr_with_weight.fit(X, y, sample_weight=sample_weight)
 
-    # check that the error is similar with 2 decimals
+    # check that the error is similar with 1 decimal
     assert (regr_no_outlier.estimator_errors_[0] ==
-            pytest.approx(regr_with_weight.estimator_errors_[0], abs=1e-2))
+            pytest.approx(regr_with_weight.estimator_errors_[0], abs=1e-1))

From 6ee41fdeb10372cde6a00e7b07d1a6725e31fcb6 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Wed, 10 Jul 2019 17:04:20 +0200
Subject: [PATCH 06/21] iter

---
 sklearn/ensemble/weight_boosting.py | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index c1864ff6e48d6..63faec2f6bf10 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -1052,12 +1052,8 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
         y_predict = estimator.predict(X)
 
         error_vect = np.abs(y_predict - y)
-
         sample_mask = sample_weight > 0
-        if np.all(~sample_mask):
-            error_max = 0
-        else:
-            error_max = error_vect[sample_mask].max()
+        error_max = error_vect[sample_mask].max()
 
         if error_max != 0:
             error_vect /= error_max

From ea46a7cac24c608ffe9b1a5954733da7f72e1591 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Tue, 10 Sep 2019 09:36:12 +0200
Subject: [PATCH 07/21] ignore line coverage

---
 sklearn/ensemble/tests/test_weight_boosting.py | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index bf5ac61d38614..182983c592456 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -499,8 +499,7 @@ def test_multidimensional_X():
 class ClassifierWithoutWeight(DummyClassifier):
     """Classifier not supporting `sample_weight`."""
     def fit(self, X, y):
-        super().fit(X, y)
-        return self
+        return super().fit(X, y)  # pragma: no cover
 
 
 @pytest.mark.parametrize("algorithm", ['SAMME', 'SAMME.R'])

From 6902effced55f33e4527ab66ed880b745c8a3f84 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Tue, 10 Sep 2019 09:44:17 +0200
Subject: [PATCH 08/21] FIX use _check_sample_weight to validate sample_weight

---
 sklearn/ensemble/weight_boosting.py | 20 ++++++--------------
 sklearn/utils/validation.py         |  4 ++--
 2 files changed, 8 insertions(+), 16 deletions(-)

diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index eece8d0117c9c..10207297917a1 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -38,6 +38,7 @@
 from ..utils.extmath import stable_cumsum
 from ..metrics import accuracy_score, r2_score
 from ..utils.validation import check_is_fitted
+from ..utils.validation import _check_sample_weight
 from ..utils.validation import has_fit_parameter
 from ..utils.validation import _num_samples
 
@@ -117,20 +118,11 @@ def fit(self, X, y, sample_weight=None):
 
         X, y = self._validate_data(X, y)
 
-        if sample_weight is None:
-            # Initialize weights to 1 / n_samples
-            sample_weight = np.empty(_num_samples(X), dtype=np.float64)
-            sample_weight[:] = 1. / _num_samples(X)
-        else:
-            sample_weight = check_array(sample_weight, ensure_2d=False)
-            # Normalize existing weights
-            sample_weight = sample_weight / sample_weight.sum(dtype=np.float64)
-
-            # Check that the sample weights sum is positive
-            if sample_weight.sum() <= 0:
-                raise ValueError(
-                    "Attempting to fit with a non-positive "
-                    "weighted number of samples.")
+        sample_weight = _check_sample_weight(sample_weight, X, np.float64)
+        sample_weight /= sample_weight.sum()
+        if sample_weight.sum() <= 0:
+            raise ValueError("Attempting to fit with a non-positive weighted "
+                             "number of samples.")
 
         # Check parameters
         self._validate_estimator()
diff --git a/sklearn/utils/validation.py b/sklearn/utils/validation.py
index 465acf48e8293..d158cd4f60ad0 100644
--- a/sklearn/utils/validation.py
+++ b/sklearn/utils/validation.py
@@ -1043,8 +1043,8 @@ def _check_sample_weight(sample_weight, X, dtype=None):
         if dtype is None:
             dtype = [np.float64, np.float32]
         sample_weight = check_array(
-                sample_weight, accept_sparse=False,
-                ensure_2d=False, dtype=dtype, order="C"
+            sample_weight, accept_sparse=False, ensure_2d=False, dtype=dtype,
+            order="C"
         )
         if sample_weight.ndim != 1:
             raise ValueError("Sample weights must be 1D array or scalar")

From a5dec847f385153e9d86b5d341536a5dbf8bfc69 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 12 Sep 2019 11:24:40 +0200
Subject: [PATCH 09/21] address jeremie comments

---
 doc/whats_new/v0.22.rst                       | 12 ++++--
 .../ensemble/tests/test_gradient_boosting.py  | 15 +------
 .../ensemble/tests/test_weight_boosting.py    | 43 +++++++++++--------
 sklearn/ensemble/weight_boosting.py           |  2 +-
 sklearn/utils/mocking.py                      | 15 +++++++
 5 files changed, 49 insertions(+), 38 deletions(-)

diff --git a/doc/whats_new/v0.22.rst b/doc/whats_new/v0.22.rst
index fa5f8ffaaade4..d9c351e09c774 100644
--- a/doc/whats_new/v0.22.rst
+++ b/doc/whats_new/v0.22.rst
@@ -176,6 +176,14 @@ Changelog
   `predict_proba` give consistent results.
   :pr:`14114` by :user:`Guillaume Lemaitre <glemaitre>`.
 
+- |Fix| :class:`ensemble.AdaBoostRegressor` where the loss should be normalized
+  by the max of the samples with non-null weights.
+  :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
+
+- |Fix| Add test for :class:`ensemble.AdaBoostClassifier` to raise consistent
+  error message when base estimator does not support `sample_weight`.
+  :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 :mod:`sklearn.feature_extraction`
 .................................
 
@@ -244,10 +252,6 @@ Changelog
  `KeyError` when using `kernel="precomputed"`.
  :pr:`14706` by :user:`Venkatachalam N <venkyyuvy>`.
 
-- |Fix| :class:`ensemble.AdaBoostRegressor` where the loss should be normalized
-  by the max of the samples with non-null weights.
-  :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
-
 :mod:`sklearn.linear_model`
 ...........................
 
diff --git a/sklearn/ensemble/tests/test_gradient_boosting.py b/sklearn/ensemble/tests/test_gradient_boosting.py
index 17e09f7f07156..d8f58f7e1ebb8 100644
--- a/sklearn/ensemble/tests/test_gradient_boosting.py
+++ b/sklearn/ensemble/tests/test_gradient_boosting.py
@@ -24,6 +24,7 @@
 from sklearn.metrics import mean_squared_error
 from sklearn.model_selection import train_test_split
 from sklearn.utils import check_random_state, tosequence
+from sklearn.utils.mocking import _NoSampleWeightWrapper
 from sklearn.utils.testing import assert_almost_equal
 from sklearn.utils.testing import assert_array_almost_equal
 from sklearn.utils.testing import assert_array_equal
@@ -1331,20 +1332,6 @@ def test_early_stopping_stratified():
         gbc.fit(X, y)
 
 
-class _NoSampleWeightWrapper(BaseEstimator):
-    def __init__(self, est):
-        self.est = est
-
-    def fit(self, X, y):
-        self.est.fit(X, y)
-
-    def predict(self, X):
-        return self.est.predict(X)
-
-    def predict_proba(self, X):
-        return self.est.predict_proba(X)
-
-
 def _make_multiclass():
     return make_classification(n_classes=3, n_clusters_per_class=1)
 
diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 182983c592456..87e17fe9a126c 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -25,6 +25,7 @@
 from sklearn.svm import SVC, SVR
 from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
 from sklearn.utils import shuffle
+from sklearn.utils.mocking import _NoSampleWeightWrapper
 from sklearn import datasets
 
 
@@ -496,16 +497,10 @@ def test_multidimensional_X():
     boost.predict(X)
 
 
-class ClassifierWithoutWeight(DummyClassifier):
-    """Classifier not supporting `sample_weight`."""
-    def fit(self, X, y):
-        return super().fit(X, y)  # pragma: no cover
-
-
 @pytest.mark.parametrize("algorithm", ['SAMME', 'SAMME.R'])
 def test_adaboostclassifier_without_sample_weight(algorithm):
     X, y = iris.data, iris.target
-    base_estimator = ClassifierWithoutWeight()
+    base_estimator = _NoSampleWeightWrapper(DummyClassifier())
     clf = AdaBoostClassifier(
         base_estimator=base_estimator, algorithm=algorithm
     )
@@ -515,32 +510,42 @@ def test_adaboostclassifier_without_sample_weight(algorithm):
         clf.fit(X, y)
 
 
-def test_adaboost_regressor_sample_weight():
+def test_adaboostregressor_sample_weight():
     # check that giving weight will have an influence on the error computed
     # for a weak learner
-    X, y = datasets.make_regression(n_features=50, random_state=0)
+    rng = np.random.RandomState(42)
+    X = np.linspace(0, 100, num=10000)
+    y = (.8 * X + 0.2) + (rng.rand(X.shape[0]) * 0.0001)
+    X = X.reshape(-1, 1)
 
-    # add an arbitrary outlier to make sure
-    X = np.vstack([X, X.sum(axis=0)])
-    y = np.hstack([y, 10])
+    # add an arbitrary outlier
+    X[-1] *= 10
+    y[-1] = 10000
 
+    # random_state=0 ensure that the underlying boostrap will use the outlier
     regr_no_outlier = AdaBoostRegressor(
-        base_estimator=LinearRegression(), n_estimators=4, random_state=0
+        base_estimator=LinearRegression(), n_estimators=1, random_state=0
     )
     regr_with_weight = clone(regr_no_outlier)
+    regr_with_outlier = clone(regr_no_outlier)
 
-    # fit 2 models:
+    # fit 3 models:
+    # - a model containing the outlier
     # - a model without the outlier
     # - a model containing the outlier but with a null sample-weight
-    # Therefore, the error of the first weak learner will be identical.
+    regr_with_outlier.fit(X, y)
     regr_no_outlier.fit(X[:-1], y[:-1])
-    sample_weight = np.array([1.] * (y.size - 1) + [0.])
+    sample_weight = np.ones_like(y)
+    sample_weight[-1] = 0
     regr_with_weight.fit(X, y, sample_weight=sample_weight)
 
-    # check that the error is similar with 1 decimal
-    assert (regr_no_outlier.estimator_errors_[0] ==
-            pytest.approx(regr_with_weight.estimator_errors_[0], abs=1e-1))
+    score_with_outlier = regr_with_outlier.score(X[:-1], y[:-1])
+    score_no_outlier = regr_no_outlier.score(X[:-1], y[:-1])
+    score_with_weight = regr_with_weight.score(X[:-1], y[:-1])
 
+    assert score_with_outlier < score_no_outlier
+    assert score_with_outlier < score_with_weight
+    assert score_no_outlier == pytest.approx(score_with_weight)
 
 @pytest.mark.parametrize("algorithm", ["SAMME", "SAMME.R"])
 def test_adaboost_consistent_predict(algorithm):
diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index 10207297917a1..820e5e6a5b984 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -928,7 +928,7 @@ class AdaBoostRegressor(RegressorMixin, BaseWeightBoosting):
     .. [1] Y. Freund, R. Schapire, "A Decision-Theoretic Generalization of
            on-Line Learning and an Application to Boosting", 1995.
 
-    .. [2] H. Drucker, "Improving Regressors using Boosting Techniques", 1997.
+    .. [2] H. Drucker, "`Improving Regressors using Boosting Techniques", 1997.
 
     """
     def __init__(self,
diff --git a/sklearn/utils/mocking.py b/sklearn/utils/mocking.py
index 45ac89f992a78..a46b8ddca8430 100644
--- a/sklearn/utils/mocking.py
+++ b/sklearn/utils/mocking.py
@@ -135,3 +135,18 @@ def score(self, X=None, Y=None):
 
     def _more_tags(self):
         return {'_skip_test': True, 'X_types': ['1dlabel']}
+
+
+class _NoSampleWeightWrapper(BaseEstimator):
+    """Wrap estimator which will not expose `sample_weight`."""
+    def __init__(self, est):
+        self.est = est
+
+    def fit(self, X, y):
+        return self.est.fit(X, y)
+
+    def predict(self, X):
+        return self.est.predict(X)
+
+    def predict_proba(self, X):
+        return self.est.predict_proba(X)

From 810c637bcd47a3a62eb4668d0225ba27119ff853 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 12 Sep 2019 11:52:25 +0200
Subject: [PATCH 10/21] fix inplace/mask copy operation

---
 sklearn/ensemble/weight_boosting.py | 17 +++++++++--------
 1 file changed, 9 insertions(+), 8 deletions(-)

diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index 820e5e6a5b984..ad4c2b5b4e279 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -1035,19 +1035,20 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
 
         error_vect = np.abs(y_predict - y)
         sample_mask = sample_weight > 0
-        error_max = error_vect[sample_mask].max()
+        masked_sample_weight = sample_weight[sample_mask]
+        masked_error_vector = error_vect[sample_mask]
 
+        error_max = masked_error_vector.max()
         if error_max != 0:
-            error_vect /= error_max
+            masked_error_vector /= error_max
 
         if self.loss == 'square':
-            error_vect **= 2
+            masked_error_vector **= 2
         elif self.loss == 'exponential':
-            error_vect = 1. - np.exp(- error_vect)
+            masked_error_vector = 1. - np.exp(-masked_error_vector)
 
         # Calculate the average loss
-        estimator_error = (sample_weight[sample_mask] *
-                           error_vect[sample_mask]).sum()
+        estimator_error = (masked_sample_weight * masked_error_vector).sum()
 
         if estimator_error <= 0:
             # Stop if fit is perfect
@@ -1066,8 +1067,8 @@ def _boost(self, iboost, X, y, sample_weight, random_state):
 
         if not iboost == self.n_estimators - 1:
             sample_weight[sample_mask] *= np.power(
-                beta,
-                (1. - error_vect[sample_mask]) * self.learning_rate)
+                beta, (1. - masked_error_vector) * self.learning_rate
+            )
 
         return sample_weight, estimator_weight, estimator_error
 

From fd50b0807514492fb5bb5986035b0047d41efa5f Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 12 Sep 2019 13:30:42 +0200
Subject: [PATCH 11/21] add default value for wrapper

---
 sklearn/utils/mocking.py | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/sklearn/utils/mocking.py b/sklearn/utils/mocking.py
index a46b8ddca8430..e98ac51347b00 100644
--- a/sklearn/utils/mocking.py
+++ b/sklearn/utils/mocking.py
@@ -1,6 +1,7 @@
 import numpy as np
 
 from ..base import BaseEstimator, ClassifierMixin
+from ..dummy import DummyClassifier
 from .validation import _num_samples, check_array
 
 
@@ -138,8 +139,14 @@ def _more_tags(self):
 
 
 class _NoSampleWeightWrapper(BaseEstimator):
-    """Wrap estimator which will not expose `sample_weight`."""
-    def __init__(self, est):
+    """Wrap estimator which will not expose `sample_weight`.
+
+    Parameters
+    ----------
+    est : estimator, default=None
+        The estimator to wrap.
+    """
+    def __init__(self, est=None):
         self.est = est
 
     def fit(self, X, y):
@@ -150,3 +157,6 @@ def predict(self, X):
 
     def predict_proba(self, X):
         return self.est.predict_proba(X)
+
+    def _more_tags(self):
+        return {'_skip_test': True}

From da5a2a0c861f174b485760e5978bf552371cc984 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 12 Sep 2019 15:08:18 +0200
Subject: [PATCH 12/21] PEP8

---
 sklearn/utils/mocking.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/sklearn/utils/mocking.py b/sklearn/utils/mocking.py
index e98ac51347b00..c89beb6083bfc 100644
--- a/sklearn/utils/mocking.py
+++ b/sklearn/utils/mocking.py
@@ -1,7 +1,6 @@
 import numpy as np
 
 from ..base import BaseEstimator, ClassifierMixin
-from ..dummy import DummyClassifier
 from .validation import _num_samples, check_array
 
 

From 1da7cc790826fb7b9b2a7cf6c2ea5d908398f67d Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 12 Sep 2019 17:36:56 +0200
Subject: [PATCH 13/21] Apply suggestions from code review

Co-Authored-By: jeremiedbb <34657725+jeremiedbb@users.noreply.github.com>
---
 doc/whats_new/v0.22.rst | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/doc/whats_new/v0.22.rst b/doc/whats_new/v0.22.rst
index e819df7b18aa4..bc2e0c6bcfe55 100644
--- a/doc/whats_new/v0.22.rst
+++ b/doc/whats_new/v0.22.rst
@@ -204,7 +204,7 @@ Changelog
   :pr:`14907` by `Adrin Jalali`_.
 
 - |Fix| :class:`ensemble.AdaBoostRegressor` where the loss should be normalized
-  by the max of the samples with non-null weights.
+  by the max of the samples with non-null weights only.
   :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
 
 - |Fix| Add test for :class:`ensemble.AdaBoostClassifier` to raise consistent

From ff2d23d81aff13f0a405143a87ca32315641b411 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 12 Sep 2019 17:38:28 +0200
Subject: [PATCH 14/21] apply jeremie comments

---
 doc/whats_new/v0.22.rst             | 4 ----
 sklearn/ensemble/weight_boosting.py | 2 +-
 2 files changed, 1 insertion(+), 5 deletions(-)

diff --git a/doc/whats_new/v0.22.rst b/doc/whats_new/v0.22.rst
index bc2e0c6bcfe55..cf9d90d78cb10 100644
--- a/doc/whats_new/v0.22.rst
+++ b/doc/whats_new/v0.22.rst
@@ -207,10 +207,6 @@ Changelog
   by the max of the samples with non-null weights only.
   :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
 
-- |Fix| Add test for :class:`ensemble.AdaBoostClassifier` to raise consistent
-  error message when base estimator does not support `sample_weight`.
-  :pr:`14294` by :user:`Guillaume Lemaitre <glemaitre>`.
-
 :mod:`sklearn.feature_extraction`
 .................................
 
diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index ad4c2b5b4e279..b7a383d35059f 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -928,7 +928,7 @@ class AdaBoostRegressor(RegressorMixin, BaseWeightBoosting):
     .. [1] Y. Freund, R. Schapire, "A Decision-Theoretic Generalization of
            on-Line Learning and an Application to Boosting", 1995.
 
-    .. [2] H. Drucker, "`Improving Regressors using Boosting Techniques", 1997.
+    .. [2] H. Drucker, "Improving Regressors using Boosting Techniques", 1997.
 
     """
     def __init__(self,

From 335b62aa84baeff96298b9fc236eb8936cc5d0d5 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Fri, 13 Sep 2019 10:08:02 +0200
Subject: [PATCH 15/21] increase coverage

---
 sklearn/ensemble/tests/test_weight_boosting.py | 18 ++++++++++++++++--
 sklearn/ensemble/weight_boosting.py            |  5 ++---
 sklearn/utils/mocking.py                       |  2 +-
 3 files changed, 19 insertions(+), 6 deletions(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 87e17fe9a126c..b89849538ff8b 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -142,9 +142,10 @@ def test_iris():
                       np.abs(clf_samme.predict_proba(iris.data) - prob_samme))
 
 
-def test_boston():
+@pytest.mark.parametrize('loss', ['linear', 'square', 'exponential'])
+def test_boston(loss):
     # Check consistency on dataset boston house prices.
-    reg = AdaBoostRegressor(random_state=0)
+    reg = AdaBoostRegressor(loss=loss, random_state=0)
     reg.fit(boston.data, boston.target)
     score = reg.score(boston.data, boston.target)
     assert score > 0.85
@@ -562,3 +563,16 @@ def test_adaboost_consistent_predict(algorithm):
         np.argmax(model.predict_proba(X_test), axis=1),
         model.predict(X_test)
     )
+
+@pytest.mark.parametrize(
+    'model, X, y',
+    [(AdaBoostClassifier(), *datasets.load_iris(return_X_y=True)),
+     (AdaBoostRegressor(), *datasets.load_boston(return_X_y=True))]
+)
+def test_adaboost_negative_weight_error(model, X, y):
+    sample_weight = np.ones_like(y)
+    sample_weight[-1] = -10
+
+    err_msg = "sample_weight cannot contain negative weight"
+    with pytest.raises(ValueError, match=err_msg):
+        model.fit(X, y, sample_weight=sample_weight)
diff --git a/sklearn/ensemble/weight_boosting.py b/sklearn/ensemble/weight_boosting.py
index b7a383d35059f..6892a39348c78 100644
--- a/sklearn/ensemble/weight_boosting.py
+++ b/sklearn/ensemble/weight_boosting.py
@@ -120,9 +120,8 @@ def fit(self, X, y, sample_weight=None):
 
         sample_weight = _check_sample_weight(sample_weight, X, np.float64)
         sample_weight /= sample_weight.sum()
-        if sample_weight.sum() <= 0:
-            raise ValueError("Attempting to fit with a non-positive weighted "
-                             "number of samples.")
+        if np.any(sample_weight < 0):
+            raise ValueError("sample_weight cannot contain negative weights")
 
         # Check parameters
         self._validate_estimator()
diff --git a/sklearn/utils/mocking.py b/sklearn/utils/mocking.py
index c89beb6083bfc..8a00466a3224d 100644
--- a/sklearn/utils/mocking.py
+++ b/sklearn/utils/mocking.py
@@ -158,4 +158,4 @@ def predict_proba(self, X):
         return self.est.predict_proba(X)
 
     def _more_tags(self):
-        return {'_skip_test': True}
+        return {'_skip_test': True}  # pragma: no cover

From 68084b5f4f574ef42540c25cc4e63670ae50977f Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Fri, 13 Sep 2019 10:18:20 +0200
Subject: [PATCH 16/21] PEP8

---
 sklearn/ensemble/tests/test_weight_boosting.py | 1 +
 1 file changed, 1 insertion(+)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index b89849538ff8b..96b9c97a63458 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -564,6 +564,7 @@ def test_adaboost_consistent_predict(algorithm):
         model.predict(X_test)
     )
 
+
 @pytest.mark.parametrize(
     'model, X, y',
     [(AdaBoostClassifier(), *datasets.load_iris(return_X_y=True)),

From 2f8054e0b065c2a1e4f473137817930b3ba241dc Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 3 Oct 2019 10:35:24 +0200
Subject: [PATCH 17/21] address adrin comments

---
 sklearn/ensemble/tests/test_weight_boosting.py | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 96b9c97a63458..4b61be7972949 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -515,7 +515,7 @@ def test_adaboostregressor_sample_weight():
     # check that giving weight will have an influence on the error computed
     # for a weak learner
     rng = np.random.RandomState(42)
-    X = np.linspace(0, 100, num=10000)
+    X = np.linspace(0, 100, num=1000)
     y = (.8 * X + 0.2) + (rng.rand(X.shape[0]) * 0.0001)
     X = X.reshape(-1, 1)
 
@@ -567,8 +567,8 @@ def test_adaboost_consistent_predict(algorithm):
 
 @pytest.mark.parametrize(
     'model, X, y',
-    [(AdaBoostClassifier(), *datasets.load_iris(return_X_y=True)),
-     (AdaBoostRegressor(), *datasets.load_boston(return_X_y=True))]
+    [(AdaBoostClassifier(), iris.data, iris.target),
+     (AdaBoostRegressor(), boston.data, boston.target)]
 )
 def test_adaboost_negative_weight_error(model, X, y):
     sample_weight = np.ones_like(y)

From cbcd68152caa9d908233f1c13b48750cb6e56d52 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 3 Oct 2019 10:40:21 +0200
Subject: [PATCH 18/21] change import for mocking

---
 sklearn/ensemble/tests/test_weight_boosting.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 4b61be7972949..440f908ba7114 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -25,7 +25,7 @@
 from sklearn.svm import SVC, SVR
 from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
 from sklearn.utils import shuffle
-from sklearn.utils.mocking import _NoSampleWeightWrapper
+from sklearn.utils._mocking import _NoSampleWeightWrapper
 from sklearn import datasets
 
 

From 362ac2abb7fdb1873307f14f13d5053e3012736b Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Thu, 3 Oct 2019 12:14:23 +0200
Subject: [PATCH 19/21] fix

---
 sklearn/ensemble/tests/test_gradient_boosting.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/sklearn/ensemble/tests/test_gradient_boosting.py b/sklearn/ensemble/tests/test_gradient_boosting.py
index b8164dd3d4a75..8398fa8471750 100644
--- a/sklearn/ensemble/tests/test_gradient_boosting.py
+++ b/sklearn/ensemble/tests/test_gradient_boosting.py
@@ -24,7 +24,7 @@
 from sklearn.metrics import mean_squared_error
 from sklearn.model_selection import train_test_split
 from sklearn.utils import check_random_state, tosequence
-from sklearn.utils.mocking import _NoSampleWeightWrapper
+from sklearn.utils._mocking import _NoSampleWeightWrapper
 from sklearn.utils.testing import assert_almost_equal
 from sklearn.utils.testing import assert_array_almost_equal
 from sklearn.utils.testing import assert_array_equal

From 31a62766580e58f3d2a593bda8dbc662cf796033 Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Tue, 22 Oct 2019 14:24:55 +0200
Subject: [PATCH 20/21] PEP8

---
 sklearn/ensemble/tests/test_weight_boosting.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 3b806f55fa868..2b82d985bc1d5 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -21,7 +21,6 @@
 from sklearn.model_selection import GridSearchCV
 from sklearn.ensemble import AdaBoostClassifier
 from sklearn.ensemble import AdaBoostRegressor
-from sklearn.ensemble import weight_boosting
 from sklearn.ensemble._weight_boosting import _samme_proba
 from scipy.sparse import csc_matrix
 from scipy.sparse import csr_matrix

From d494b3ffad4ad2a5333907e0541fe0cf0b84bf2b Mon Sep 17 00:00:00 2001
From: Guillaume Lemaitre <g.lemaitre58@gmail.com>
Date: Wed, 23 Oct 2019 16:22:47 +0200
Subject: [PATCH 21/21] address comment adrin

---
 sklearn/ensemble/tests/test_gradient_boosting.py | 4 ++--
 sklearn/ensemble/tests/test_weight_boosting.py   | 4 ++--
 sklearn/utils/_mocking.py                        | 2 +-
 3 files changed, 5 insertions(+), 5 deletions(-)

diff --git a/sklearn/ensemble/tests/test_gradient_boosting.py b/sklearn/ensemble/tests/test_gradient_boosting.py
index 8398fa8471750..f19c2cc09ce5e 100644
--- a/sklearn/ensemble/tests/test_gradient_boosting.py
+++ b/sklearn/ensemble/tests/test_gradient_boosting.py
@@ -24,7 +24,7 @@
 from sklearn.metrics import mean_squared_error
 from sklearn.model_selection import train_test_split
 from sklearn.utils import check_random_state, tosequence
-from sklearn.utils._mocking import _NoSampleWeightWrapper
+from sklearn.utils._mocking import NoSampleWeightWrapper
 from sklearn.utils.testing import assert_almost_equal
 from sklearn.utils.testing import assert_array_almost_equal
 from sklearn.utils.testing import assert_array_equal
@@ -1317,7 +1317,7 @@ def test_gradient_boosting_with_init(gb, dataset_maker, init_estimator):
     gb(init=init_est).fit(X, y, sample_weight=sample_weight)
 
     # init does not support sample weights
-    init_est = _NoSampleWeightWrapper(init_estimator())
+    init_est = NoSampleWeightWrapper(init_estimator())
     gb(init=init_est).fit(X, y)  # ok no sample weights
     with pytest.raises(ValueError,
                        match="estimator.*does not support sample weights"):
diff --git a/sklearn/ensemble/tests/test_weight_boosting.py b/sklearn/ensemble/tests/test_weight_boosting.py
index 440f908ba7114..ada5cab0d2aeb 100755
--- a/sklearn/ensemble/tests/test_weight_boosting.py
+++ b/sklearn/ensemble/tests/test_weight_boosting.py
@@ -25,7 +25,7 @@
 from sklearn.svm import SVC, SVR
 from sklearn.tree import DecisionTreeClassifier, DecisionTreeRegressor
 from sklearn.utils import shuffle
-from sklearn.utils._mocking import _NoSampleWeightWrapper
+from sklearn.utils._mocking import NoSampleWeightWrapper
 from sklearn import datasets
 
 
@@ -501,7 +501,7 @@ def test_multidimensional_X():
 @pytest.mark.parametrize("algorithm", ['SAMME', 'SAMME.R'])
 def test_adaboostclassifier_without_sample_weight(algorithm):
     X, y = iris.data, iris.target
-    base_estimator = _NoSampleWeightWrapper(DummyClassifier())
+    base_estimator = NoSampleWeightWrapper(DummyClassifier())
     clf = AdaBoostClassifier(
         base_estimator=base_estimator, algorithm=algorithm
     )
diff --git a/sklearn/utils/_mocking.py b/sklearn/utils/_mocking.py
index fb6b410720622..3edcf8da53a95 100644
--- a/sklearn/utils/_mocking.py
+++ b/sklearn/utils/_mocking.py
@@ -137,7 +137,7 @@ def _more_tags(self):
         return {'_skip_test': True, 'X_types': ['1dlabel']}
 
 
-class _NoSampleWeightWrapper(BaseEstimator):
+class NoSampleWeightWrapper(BaseEstimator):
     """Wrap estimator which will not expose `sample_weight`.
 
     Parameters